In recent years, there has been studied a data input technique of inputting data by a gesture (motion) or a posture (pose) of a user as an input device for inputting data into a personal computer or the like, in addition to a mouse, a graphics tablet, and a touch pad.
Since data is input by, for example, a gesture or a posture of a hand of a user in the data input technique, it is necessary to precisely extract the shape or movement of the hand of the user from a captured image obtained by capturing an image of the user.
As extraction techniques for extracting the shape of the hand of the user, there are a pattern matching method using pattern matching of images, a skin area extraction method of extracting a skin area of the user, and the like.
In the pattern matching method, for example, a plurality of shape images obtained by capturing images of hands having various shapes and sizes are learned in advance, and a shape of a hand represented in a shape image that is most similar to a captured image (for example, shape image having a minimum sum of differences between pixel values of corresponding pixels) is extracted as the shape of the hand of the user.
In the pattern matching method, however, in a case where a captured image is obtained under conditions that are different from those when shape images are captured (for example, image capturing direction, level of illumination, background, size of subject when image thereof is captured, and the like), it may be impossible to precisely extract the shape of the hand of the user.
Particularly when a shape of a hand within a captured image is largely different from shapes of hands within shape images or when the hand within the captured image overlaps a face or the like in the case where the shape of the hand is extracted, the shape of the hand is difficult to be precisely extracted as compared to, for example, a case where a shape of a face is extracted.
In addition, when the shape of the hand is required to be extracted in real-time, the pattern matching needs a huge amount of calculations, which causes a problem in many cases.
Next, in the skin area extraction method, a skin area representing a skin of the user within the captured image is extracted using skin color information representing colors of the human skin.
However, in the skin area extraction method using the skin color information, it is difficult to distinguish colors of the skin from colors close thereto. Further, since colors of the skin are different between races, it is impossible to appropriately extract skin areas for all races.
In this regard, there is proposed recently an extraction technique in which spectral reflectance characteristics are used for extracting a skin area within a captured image using the fact that a reflectance change of the skin with respect to a wavelength is constant irrespective of a race (see, for example, Non-patent Document 1).    Non-patent Document 1: Yasuhiro Suzuki et al., “Detection Method of Skin Region by Near-IR Spectrum Multi-Band”, IEEJ Transactions on Electronics, Information and Systems Vol. 127, No. 4, 2007, Japan